MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
• The information required by the mean-variance approach is substantial when the number n of assets is
large; there are n mean values, n variances, and n(n − 1)/2 covariances - a total of 2n + n(n − 1)/2
parameters.
• Single-factor model: Suppose that there are n assets, indexed by i, with rates of return ri , i = 1, 2, . . . , n.
Suppose then that there is a single factor f , a random quantity, and assume that the rates of return and
the factor are related by
ri = ai + bi f + ei , (1)
for i = 1, . . . , n. In Equation (1), ai and bi are constants and ei is a random quantity that represents the
error of the model. bi is termed the factor loading that measures the sensitivity of the return to the factor.
For the error we assume E[ei ] = 0 ∀i, because any non-zero mean could be transferred to ai . Moreover, we
assume that the errors are uncorrelated with f and with each other; that is,
Cov[ei , ej ] = E[(ei − ēi )(ej − ēj )] = E[ei ej ] = 0 ∀i 6= j
Cov[f, ei ] = E[(f − f¯)(ei − ēi )] = E[(f − f¯)ei ] = 0 ∀i.
The variances of the error terms, denoted by σe2i , are assumed known.
• If we agree to use a single-factor model, the standard parameters for mean-variance analysis can be deter-
mined as
r̄i = ai + bi f¯
σi2 = b2i σf2 + σe2i
σij = bi bj σf2 , i 6= j
Cov[ri , f ]
bi = .
σf2
The only parameters to be estimated for this model are ai ’s, bi ’s, σe2i ’s, and f¯ and σf2 ’s - a total of just
3n + 2 parameters.
• Suppose that there are n assets with rates of return governed by theP single-factor model (1), and suppose
that a portfolio of these assets is constructed with weights wi , with ni=1 wi = 1. Then the rate of return
r of the portfolio is
X n n
X n
X n
X
r= wi ri = wi a i + wi bi f + wi e i ,
i=1 i=1 i=1 i=1
Pn Pn Pn
and denoting a = i=1 wi ai , b= i=1 wi bi and e = i=1 wi ei , we can write the above formula as
r = a + bf + e.
Similarly to the single asset case, a and b are constants and e is a random variable. Because E[ei ej ] =
0, i 6= j, the variance of e is
! n
n
" n # n
X X X X
2 2 2 2
σe = E[e ] = E wi ei wj ej = E wi e i = wi2 σe2i , (2)
i=1 j=1 i=1 i=1
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
and because Cov[f, ei ] = 0 ∀i ⇒ Cov[f, e] = 0, the overall variance of the portfolio is
σ 2 = Var[a + bf + e] = b2 Var[f ] + 2bCov[f, e] + Var[e] = b2 σf2 + σe2 . (3)
• The CAPM can be interpreted as a special case of a single-factor model. Suppose we model the excess
rates of return of stocks ri − rf (where rf is the risk-free rate) with a single-factor model, with the factor
being the excess rate of return of the market rM − rf . Then the factor model becomes
ri − rf = αi + βi (rM − rf ) + ei .
• Arbitrage Pricing Theory (APT) is an alternative theory of asset pricing that can be derived from the
factor model framework. Instead of the strong equilibrium assumption of the CAPM theory, APT assumes
that
(i) when returns are certain, investors prefer greater return to lesser return,
(ii) the universe of assets being considered is large, and
(iii) there are no arbitrage opportunities.
Specifically, APT assumes that all asset rates of return satisfy a single-factor model with no error term,
and hence the uncertainty with a return is due only to the uncertainty in the factor f . For assets i and j,
we write
ri = ai + bi f rj = aj + bj f. (4)
We then select w such that the coefficient of f in this equation is zero; that is,
bj
r = w(ai + bi f ) + (1 − w)(aj + bj f ) = wai + (1 − w)aj + (wbi + (1 − w)bj ) f ⇒ w =
| {z } bj − bi
=0
If there is a risk-free asset rf , the return of this portfolio must have this same rate, and even if there is no
explicit risk-free asset, all portfolios constructed this way must have the same rate of return - otherwise
there would be an arbitrage opportunity. We denote this rate by λ0 and find that
r = wai + (1 − w)aj = λ0
bj ai bi aj
⇒ − = λ0
bj − bi bj − bi
⇒λ0 (bj − bi ) = ai bj − aj bi
aj − λ0 ai − λ0
⇒ = = c. (5)
bj bi
The last equation holds for all i and j (because we could select any assets i and j that satisfy (??)), which
is why it must be a constant c. We use this information to write a simple formula for the expected rate of
return of asset i as follows:
r̄i = ai + bi f¯ = λ0 + bi c + bi f¯ = λ0 + bi λ1 ,
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
where λ1 = c + f¯ is a constant (same for all assets i). This statement can be generalized for multiple
factors:
m
X
ri = ai + bij fj
j=1
m
X
⇒ r̄i = λ0 + bij λj ,
j=1
where the value λj is the price of risk associated with the factor fj , often called the factor price.
• Value = how much a certain event is appreciated
Utility = how much an uncertain event is appreciated, when making choices associated with risk
These are concepts of decision theory, which is more extensively studied in course MS-E2134 - Decision
making and problem solving.
• A utility function U (x) relates possible wealth levels x to a real value. Once a utility function is defined,
all alternative random wealth levels are ranked by evaluating their expected utility values, that is,
E[U (x)] > E[U (y)] ⇔ alternative x is preferred to alternative y.
• Decision theory assumes non-satiation, which reflects the idea that, every thing else being equal, investors
always want more money. Moreover, it is usually assumed that investors are risk averse, that is, investments
with the smallest standard deviation for the given mean are preferred. Non-satiation assumption leads to
an increasing utility function, and the utility function of a risk averse investor is concave.
• The degree of risk aversion exhibited by a utility function can be formally defined by the Arrow-Pratt
absolute risk aversion coefficient, which is
U 00 (x)
a(x) = − ,
U 0 (x)
where the denominator U 0 (x) normalizes the coefficient. For many individuals, risk aversion decreases as
their wealth increases, and consequently, a(x) is often decreasing.
• The certainty equivalent CE of a random wealth variable x is defined to be the amount of a certain wealth
that has a utility level equal to the expected utility of x, that is,
U (CE) = E[U (x)]
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
7.1 (L8.1) (A simple portfolio) Someone who believes that the collection of all stocks satisfies a single-factor
model with the market portfolio serving as the factor gives you information on three stocks which make up
a portfolio. (See Table 1.) In addition, you know that the market portfolio has an expected rate of return
of 12% and a standard deviation of 18%. The risk-free rate is 5%.
a) What is the portfolio’s expected rate of return?
b) Assuming the factor model is accurate, what is the standard deviation of this rate of return?
Table 1: Simple Portfolio.
Stock Beta Standard deviation of random error term Weight in portfolio
A 1.10 7.0% 20%
B 0.80 2.3% 50%
C 1.00 1.0% 30%
Solution:
rf = 5% rM = 12% σM = 18%
βi σei wi
A 1.10 7.0% 20%
B 0.80 2.3% 50%
C 1.00 1.0% 30%
a) The beta of the portfolio is a weighted combination of the individual betas:
βp = wA βA + wB βB + wC βC = 0.92
Hence, applying the CAPM to the portfolio we find
r̄p = rf + βp (r̄M − rf ) = 0.05 + 0.92(0.12 − 0.05) = 0.1144 = 11.44%.
b) Using formula (??) The variance of the error terms of the portfolio is
n
X
σe2 = wi2 σe2i = 0.22 × 0.072 + 0.52 × 0.0232 + 0.32 × 0.012 = 0.00033725
i=1
Then, using formula (??), the overall variance of the portfolio is
σ 2 = b2 σf2 + σe2 = βp2 σf2 + σe2 = 0.922 × 0.182 + 0.00033725 = 0.02776,
and the standard deviation of the portfolio is
√
σ = σ 2 = 0.167 = 16.7%.
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
7.2 (L8.2) (APT factors) Two stock are believed to satisfy the two-factor model
r1 = a1 + 2f1 + f2
r2 = a2 + 3f1 + 4f2 .
In addition, there is a risk-free asset with a rate of return rf = 10%. It is known that r̄1 = 15% and r̄2 =
20%. What are the values of λ0 , λ1 and λ2 for this model?
Solution:
When the rates of return of the stocks depend on two factors as
ri = ai + bi1 f1 + bi2 f2 ,
the expected rate of return any asset i is
r̄i = λ0 + bi1 λ1 + bi2 λ2 .
Because there is a risk-free asset,
λ0 = rf = 10%.
Because we know the expected rates of return r̄i of stocks 1 and 2, we can solve λ1 and λ2 from the equation
system
15% = 10% + 2λ1 + λ2
20% = 10% + 3λ1 + 4λ2 .
Solving this yields
λ1 = 2% λ2 = 1%.
Hence the prices of risks of factors 1 and 2 are 2% and 1%, respectively.
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
7.3 (L8.4) (Variance estimate) Let ri , for i = 1, 2, . . . , n, be independent samples of a return r of mean r̄ and
variance σ 2 . Define the estimates
n
1X
ˆ
r̄ = ri
n
i=1
n
1 X
s2 = (ri − r̄ˆ)2 .
n−1
i=1
Show that E[s2 ] = σ2.
Solution:
We show that s2 is an unbiased estimate of the variance. First we write the formula of the average r̄ˆ in
the formula of the sample variance. Then we add and subtract the true expected rate of return r̄ inside
the squared term. These modifications yield
2 2
n n n n n
" #
1 X 1 X 1 X 1 X 1 X
E[s2 ] = E (ri − r̄ˆ)2 = E ri − rj = E (ri − r̄) − (rj − r̄) .
n−1 n−1 n n−1 n
i=1 i=1 j=1 i=1 j=1
Extracting ri from the inner summation and moving the factor 1/(n − 1) outside the expected value yields
2
n
1 X
1 − 1 (ri − r̄) − 1
X
E (rj − r̄) .
n−1 n n
i=1 j6=i
We then expand the square inside the summation to get
2
n 2
1 X 1 1 1 X 1 X
E 1− (ri − r̄)2 − 2 1 − (ri − r̄) (rj − r̄) + (rj − r̄) .
n−1 n n n n
i=1 j6=i j6=i
h P i2
Moving the expected value inside the summation and expanding the term (1/n) j6=i (rj − r̄) yields
n 2 X
1 X
1 − 1 E (ri − r̄)2 − 2 1 − 1 1
1 X X
E [(ri − r̄)(rj − r̄)] + 2 E [(rj − r̄)(rk − r̄)] .
n−1 n n n n
i=1 j6=i j6=i k6=i
The samples were assumedindependent and hence have zero covariance, that is, E [(ri − r̄)(rj − r̄)] = σij =
0, i 6= j. Note also that E (ri − r̄)2 = σ 2 . These modifications give the above formula as
n
" # " #
1 2 2
2
1 X 1 n 1 1
1− σ + 2 (n − 1)σ 2 = σ 2 1− + 2 (n − 1) .
n−1 n n n−1 n n
i=1
This can be simplified into
2 n 2 1 1 1 2 n 1 n−1
σ 1− + 2 + − 2 =σ 1− = σ2 = σ2
n−1 n n n n n−1 n n−1
Thus, we have shown that E[s2 ] = σ 2 .
Note that if the true expected
value
P r̄ was known, the denominator
of the variance estimate would be n
instead of n − 1, because E (1/n) ni=1 (ri − r̄)2 = (1/n) ni=1 E (ri − r̄)2 = (1/n)nσ 2 = σ 2 .
P
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
7.4 (L8.7) (Clever, but no cigar) Kalle Virtanen figured out a clever way to get 24 samples of monthly returns
in just over one year instead of only 12 samples; he takes overlapping samples; that is, the first sample
covers Jan 1 to Feb 1, and the second sample covers Jan 15 to Feb 15, and so forth. He figures that the
error in his estimate of r̄, the mean monthly return, will be reduced by this method. Analyse Kalle’s idea.
How does the variance of his estimate compare with that of the usual method of using 12 non-overlapping
monthly returns?
Solution:
First divide the year into half-month intervals and index these time points by i. Let ri be the return over
the i-th full month (but some will start midway through the month). We let r̄m and σm2 denote the monthly
expected return and variance of that return.
Now let ρi be the return over the i-th half-month period. Assume that these returns are uncorrelated.
The return over any monthly period is a sum of two half-month returns; that is, the monthly return ri is
ri = ρi + ρi+1 . It is easy to see that
r̄i = ρ̄i + ρ̄i+1 ⇒ r̄m = 2ρ̄ ⇒ ρ̄ = r̄m /2
2
σm = Var[ρi + ρi+1 ] = Var[ρi ] + Var[ρi+1 ] = 2σρ2 ⇒ σρ2 = σ 2 /2,
where ρ̄ is the expected half-monthly return and σρ2 is the variance of that return. The covariance of two
monthly returns ri and rj is then
Cov[ri , rj ] = Cov[ρi + ρi+1 , ρj + ρj+1 ]
= Cov[ρi , ρj ] + Cov[ρi , ρj+1 ] + Cov[ρi+1 , ρj ] + Cov[ρi+1 , ρj+1 ]
2
Var[ri ] = σ ,
if i = j
2
⇒ Cov[ri , rj ] = Var[ρi ] = σ /2, if |i − j| = 1
0, otherwise.
Now for Kalle’s scheme we form the estimate
24
1 X
r̄ˆ = ri .
24
i=1
We need to evaluate
24 24 24
" #
1 X 1 XX
Var r̄ˆ = Var ri = 2 Cov [ri , rj ]
24 24
i=1 i=1 j=1
23
1 X
= 2 Var[r1 ] + Cov[r1 , r2 ] + (Cov[rj−1 , rj ] + Var[rj ] + Cov[rj , rj+1 ]) + Cov[r23 , r24 ] + Var[r24 ]
24
j=2
Hence we have
1 1 2 × 24 2 1 1 1
Var r̄ˆ = 2 24σ 2 + 2 × 23σ 2 /2 = 2 24σ 2 + 24σ 2 − σ 2 = σ − 2 σ2 = σ2 − 2 σ2.
24 24 24 2 24 12 24
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
For twelve non-overlapping months of data we would have
12
" #
1 X 12 1
Var[r̄ˆ] = Var ri = 2 σ 2 = σ 2 .
12 12 12
i=1
The difference σ 2 /242 ≈ 0.0017σ 2 is caused by the longer estimation period for the half-month case; the
last sample is actually partly from the following year, because sample r24 covers Dec 15 to Jan 15. Hence
the estimation precision cannot be increased by sampling with half-month intervals; the gain of greater
sample size is lost because of the correlation of the overlapping samples.
MS-E2114 Investment science
Leevi Olander, Jaakko Wallenius Exercise 7 Solutions
7.5 (L9.1) (Certainty equivalent) An investor has utility function U (x) = x1/4 for salary. He has a new job offer
which pays 80000 e with a bonus. The bonus will be 0 e, 10000 e, 20000 e, 30000 e, 40000 e, 50000 e, or
60000 e, each with equal probability. What is the certainty equivalent of this job offer?
Solution:
The certainty equivalent can be solved by taking the inverse utility function from the expected utility of a
random wealth variable, because
U (CE[X]) = E [U (X)] ⇒ CE[X] = U −1 (E[U (X)])
U (x) = x1/4 ⇒ U −1 (x) = x4
The possible incomes and their utility levels (found by taking the 1/4-th power) are
Bonus Total income U (x) p(x) p(x)U (x)
0 80000 16.82 1/7 2.40
10000 90000 17.32 1/7 2.47
20000 100000 17.78 1/7 2.54
30000 110000 18.21 1/7 2.60
40000 120000 18.61 1/7 2.66
50000 130000 18.99 1/7 2.71
60000 140000 19.34 1/7 2.76
P
18.15
Hence
E [U (X)] = 18.15 ⇒ CE[X] = U −1 (E[U (X)]) = 18.154 = 108610e.
